Abnormal RNA splicing of stathmin-2 and UNC13A is a key feature of Alzheimer's disease linked to TDP-43 protein

Total RNA was extracted from 20 to 40 mg of frozen post-mortem brain tissue, or neuroblastoma SH-SY5Y cells, and lysed using Trizol reagent (ThermoFisher, Cat# 10,296,010). Post-mortem frozen brain samples were obtained from the Massachusetts Alzheimer’s Disease Research Center Neuropathology Core operated under the Massachusetts General Hospital IRB guidelines. A mechanical tissue homogenizer was used for brain tissue. RNA was then purified using Phase Lock Gel Heavy tubes (QuantaBio, Cat# 10847-802) and Isolate RNA Mini Kit (Bioline, Cat#BIO-52073). RNA concentration and quality were measured using nanodrop and 750–1000 ng of total RNA used for cDNA synthesis, using High-Capacity cDNA Reverse Transcription Kit (ThermoFisher, Cat# 4374966) according to the manufacturer’s instructions.

Reverse transcription polymerase chain reactions (RT-PCRs) were performed using Q5 High-Fidelity DNA polymerase (New England BioLabs, Cat# M0492S) according to the manufacturer’s instructions. Quantitative real-time PCR (qRT-PCR) was performed using iTaq Universal SYBR Green (Bio-rad, Cat#1725121), or Taq™ Universal Probes Supermix (Bio-rad, Cat#1725131). In order to increase specificity of the signal, an optimized combination of primers and probes were used for both STMN2 and UNC13A cryptic exons. Probes and primers used for these reactions are listed in Supplementary Table 1. Reactions were run in triplicates in CFX96 real-time PCR machine (Biorad). For analysis, target RNA levels were normalized to the RNA levels of GAPDH (reference transcript) and to Alzheimer’s disease patients. As STMN2 and UNC13A cryptic exons are normally not expressed, certain samples did not display any amplification of these aberrant transcripts. A sample was considered positive for cryptic exons when there was product amplification in all 3 triplicates evaluated.

SH-SY5Y human neuroblastoma cells (ATCC Cat# CRL-2266) were cultured in DMEM F12 (Gibco, Cat#1132-033) containing 1% of Penicillin–Streptomycin (Gibco, Cat#15140-122) and 10% of Fetal Bovine Serum (Sigma, Cat# F0926) and kept at 37 °C and 5% CO₂. TDP-43 knockdown was achieved by transfecting cells with pooled siRNAs against TARDBP (ON-TARGETplus siRNA, Dharmacon, Cat# L-012394) or control siRNA (ON-TARGETplus Non-targeting Control Pool, Dharmacon, Cat# D001810–10) using Lipofectamine RNAiMAX Transfection Reagent (Cat# 13778075) in Opti-MEM™ (Gibco, Cat#31985070). For the dose-dependency analyses we used decreasing amounts of siRNAs ranging from 25 to 0.125 pmol per well. Cells were treated with siRNA either for 3 or 5 days, as previous studies reported decrease in UNC13A transcripts after longer knockdown of TDP-43 [16, 57].

Cells containing the N352S mutation in both TDP-43 alleles were previously reported [62]. In short, a plasmid containing a single guide RNA (GCGGGTAATAACCAAAACCA) was electroporated in SH-SY5Y cells using a Amaxa Nucleofector (assay A-023), along with a 180-nucleotide long single-stranded donor oligonucleotides (IDT), containing the desired ALS-causing mutation and four synonymous single-nucleotide replacements to avoid DNA cleavage recurrence by Cas9. Cells were sorted and single-cell seeded into a 96-well plate to obtain individual clones. Replacement of AAT to AGT (c.1055A > G) was confirmed by Sanger sequencing.

To obtain cells overexpressing heterogeneous nuclear ribonucleoprotein L (hnRNP L), SH-SY5Y cells were treated with polybrene (Santa Cruz, Cat# sc-134220) and transduced with lentiviral particles encoding the HNRNP L transcript (Origene Cat# CW307794V) or a GFP control (Origene Cat# CW307797V). Transduced cells were then selected using 1 μg/mL puromycin (ThermoFisher, Cat#A1113803). Overexpression of hnRNP L was verified using qRT-PCR (Forward 5′-GTGGAGATGGCTGATGGCTAC-3′, Reverse 5′ GCTCATCGCAGATCTCAAAGAAG-3′) and Western blot (Antibody Sigma Cat#R4903) with GFP overexpressing cells used as control. SH-SY5Y cells were transfected using lipofectamine with 2.5 pMol of TDP-43 or control siRNAs and harvested 3 days later for qRT-PCR analysis.

Cell toxicity was assessed using the CellTox Green Toxicity assay (Promega, Cat#G8742) according to manufacturer’s instructions. Cells were seeded in a 96-well plate, and toxicity measured after 5 days of siRNA treatment. Cells were treated with CellTox Green Reagent for 15 min at room temperature and shielded from light. As positive control, the cells were lysed for 30 min using the lysis buffer provided by the Promega. Fluorescence was measured on a Synergy2 plate reader (BioTek Instruments) using 485 nm excitation and 528 nm emission.

Formalin fixed paraffin-embedded brain sections from control or Alzheimer’s disease patients were obtained from the Massachusetts Alzheimer’s Disease Research Center Neuropathology Core operated under the Massachusetts General Hospital IRB guidelines. Brains donated to the Massachusetts Alzheimer’s Disease Research Center, are cut in half, one hemisphere is fresh frozen while the contralateral hemisphere is fixed in 10% neutral buffered formalin. The fixed hemisphere was sectioned and an average of 19–22 distinct regions of interest were mounted in blocks for histological examination. Alzheimer’s disease patients were classified using clinical and neuropathological criteria [30], and aged and sex matched control subjects were chosen that had no significant clinical or neuropathological abnormalities. Demographic characteristics of Alzheimer’s disease patients and control individuals selected are described in Supplementary Table 2. The cohort of patients in this study did not include any patient with pure LATE-NC (i.e. without Alzheimer’s disease neuropathological changes).

For immunohistochemistry and immunofluorescence of TDP-43, sections were first dehydrated in an oven for 1 h at 70 °C, followed by sequential steps of deparaffinization and rehydration using Citrosolv (Decon Laboratories, Cat# 04-355-121), 100% Ethanol (Fisher Scientific, Cat# BP28184), 90% Ethanol, 70% Ethanol and demineralized water. Antigen retrieval was performed using Sodium Citrate Buffer (Fisher Scientific, Cat#S279-500) pH 6.0 for 20 min at 120 °C. Endogenous peroxidases were quenched with a treatment of 3% H₂O₂ (Millipore Sigma, Cat# HX0635) for 30 min. Sections were blocked for one hour in a solution of 1:10 donkey serum (Jackson Immuno Research Labs, Cat#NC9624464) in TBS-0.1% Triton (Tris Buffered Saline, Boston BioProducts, Cat# BM-301; Triton Sigma, Cat#T8787). Sections were incubated overnight at 4 °C with phosphorylated TDP-43 primary antibody (CosmoBio Mouse Polyclonal, Cat#CAC-TIP-PTD-M01, 1:10,000), or TDP-43 primary antibody (Proteintech, Rabbit Polyclonal, Cat# 10782-2-AP, 1:500) diluted in blocking solution. The next day sections were washed 3× using TBS-0.1% Triton, and then incubated with corresponding secondary antibodies for 2 h at room temperature. For immunohistochemistry the EnVision + Single Reagent HRP Mouse (Agilent, Cat# K400111-2) system was used according to manufacturer’s instructions, and cell nuclei were stained for 30 s using Hematoxylin (Fisher Scientific, Cat#3530-32). For immunofluorescence, slides were incubated with Alexa Fluor anti-rabbit 488 and Alexa Fluor anti-mouse 568 secondary antibodies, and cell nuclei were stained using DAPI. Finally, slides were dehydrated by a sequential treatment with demineralized water, 70% Ethanol, 90% Ethanol, 100% Ethanol and Citrosolv, and then coverslipped with Epredia mounting media (Epredia, Cat#4112). At the end, immunofluorescent slides were treated with 0.1% Sudan Black, to quench the human tissue auto-fluorescence. Total tau (1:6000 dilution, Agilent/Dako, #A002401-2, 1), and amyloid-β 6F/3D (1:600 dilution, Agilent/Dako, #M087201-2) were stained using a Leica BOND-RX automated stainer.

For immunohistochemistry, whole-slide images were taken using brightfield 40× objective (NanoZoomer, Hamamatsu), and a confocal Nikon C2 Confocal was used to image slides stained by immunofluorescence at a 60× magnification. To quantify the percentage of area occupied by phosphorylated TDP-43, Tau or amyloid-β, five images of the same area were selected within the amygdala for each slide and extracted for quantification using FIJI [82]. The Color Threshold and Analyze particles plugins from FIJI were used to quantify the area stained for each protein. Investigators were blinded to the disease status during the experimental procedure and analysis.

TDP-43 staging was determined according to the simplified staging method proposed by Nelson et al. [67]. In short, patients were classified according to the presence of TDP-43 proteinopathy in the amygdala (stage I), hippocampus (stage II) and frontal cortex (stage III). The information obtained for each individual is included in Supplementary Table 3.

The Mayo Clinic RNA-seq cohort has been made available to the research community and described in detail previously [2, 3, 98]. In brief, Alzheimer’s disease and control subjects were diagnosed neuropathologically at autopsy [61]. Alzheimer’s disease subjects from the Mayo Clinic Brain Bank had definite neuropathologic diagnosis according to the National Institute of Neurological and Communicative Disorders and Stroke and the Alzheimer’s Disease and Related Disorders Association (NINCDS–ADRDA) criteria and had Braak neurofibrillary tangle (NFT) stage ≥ 4.0. Control subjects, either from the Mayo Clinic Brain Bank or the Banner Sun Health Institute, had Braak NFT stage of 3.0 or less, Consortium to Establish a Registry for Alzheimer’s Disease (CERAD) neuritic and cortical plaque densities of 0 (none) or 1 (sparse) and lacked neuropathologic diagnoses for neurodegenerative diseases. Temporal cortex samples underwent RNA extractions via the Trizol/chloroform/ethanol method, followed by DNase and clean-up of RNA using Qiagen RNeasy Mini Kit and Qiagen RNase-Free DNase Set. The quantity and quality of RNA samples were determined by the Agilent 2100 Bioanalyzer using the Agilent RNA 6000 Nano Chip. Samples included in this study all have RIN ≥ 5.0. After quality control procedures, final samples included in this study consisted of 80 Alzheimer’s disease patients and 69 controls. In a subset of 38 Alzheimer’s disease samples (18 TDP-43 positive and 20 negative) the presence of TDP-43 inclusion bodies was determined by immunohistochemistry with antibodies directed against pathological TDP-43 [5].

RNA sequencing Library preparation and sequencing of the samples were conducted at the Mayo Clinic Genome Analysis Core using TruSeq RNA Sample Prep Kit (Illumina, San Diego, CA). The library was sequenced on Illumina HiSeq2000 instruments, generating 101 base-pair, paired-end raw reads. Raw reads were processed through MAPR-Seq pipeline v1.0 [39]. Reads of low base-calling Phred scores were removed, the remaining reads were aligned to the human reference genome build hg19 using Tophat v2.0.12 [91], and were counted for genes using Subread 1.4.4 [51]. Quality control measures were obtained from both pre-alignment and post-alignment reads using RSeQC toolkit and fastQC (https://www.bioinformatics.babraham.ac.uk/projects/fastqc/↗). Samples that had high RNA degradation, low reads mappability, or inconsistency between recorded sex and estimated sex using RNA-seq chromosome Y expression were removed from downstream analysis. Additional information is also available in our previous publications [2, 3, 98].

The Mount Sinai Brain Bank (MSBB) RNA-seq data was downloaded from the Accelerating Medicines Partnership–Alzheimer’s Disease (AMP–AD) Knowledge Portal (syn3157743) as previously described [97]. For the MSBB RNA-seq data, pre-mortem clinical diagnosis was used to categorize Alzheimer’s disease patients and controls. We used trimmed mean of M-values (TMM) provided by the MSBB study which were already adjusted for batch, sex, race, age at death, PMI, RNA integrity number, exonic rate and ribosomal RNA rate.

For both RNA-seq datasets, reads of full-length STMN2 (chr8: 80,523,049–80,578,410 in hg19), UNC13A (chr19: 17,712,145–17,799,163 in hg19), the cryptic exon of STMN2 (chr8: 80,529,057–80,529,284 in hg19) and the cryptic exon of UNC13A (chr19: 17,753,223–17,753,350 in hg19) were counted for each sample using the bedtools program (v2.27.1) [77] with at least 75% overlap in the region of interest. These counts were then normalized by the total number of reads of all genes (counts per million reads, CPM), followed by Log2 transformation to obtain the expression levels in Log2(CPM). Plots were created using the ggplot2 package in R and p-values were obtained using the Wilcoxon rank-sum test. Chi-Square tests were performed using the chisq.test function and the plots were created using the ggplots package (v3.36) in R. In the plot of stratified Braak for the Mayo Clinic cohort, non-integer scores (*0.5) were rounded to the previous integer, leading to 69, 6, 34 and 40 subjects with scores 0–III, IV, V and VI, respectively. The Braak stages 0 to III were considered controls and pooled together for the analysis. Information about the presence or absence of TDP-43 was available only for 38 cases, most of them with Braak stages of V (4 TDP-43 positive and 9 negative) or VI (13 TDP-43 positive and 10 negative).

Individual-nucleotide resolution cross linking and immunoprecipitation (iCLIP) for TDP-43 in SH-SY5Y cells (ERR039854) previously generated by Tollervey et al. [90] was mapped to the human genome GRCh38/hg38 using star/2.7.9a. Sorted bam files were visualized and plotted using sashimi plot function in IGV (version 2.12.3) for the respective genes of interest (UNC13A and STMN2).

Statistical analysis was performed using GraphPad Prism 8.0 software. Normality was tested using D’Agostino & Pearson test. When all groups were normally distributed, differences between groups were tested using an unpaired t-tests (for comparison of two groups) or one-way ANOVA, followed by Holm-Šídák’s Multiple comparisons post-hoc test (for comparison of more than two groups). When at least one group was not normally distributed, differences between groups were tested using Mann–Whitney test (for comparison of two groups) or Kruskal–Wallis, followed by a Dunn’s Multiple comparisons post-hoc test (for comparison of more than two groups). Correlations were tested using the Spearman’s rank correlation coefficient. A p-value of ≤ 0.05 was considered statistically significant. The following code was used to indicate the level of significance: * ≤ 0.05; ** ≤ 0.01; *** ≤ 0.001.

The data from this study are available from the corresponding authors, upon request. Mayo Clinic brain RNA-seq data are available from the AD Knowledge Portal (syn5550404). MSBB RNA-seq data are available from the AD Knowledge Portal (syn3157743).

Both STMN2 and UNC13A RNAs are mis-spliced upon reduction of nuclear TDP-43 with incorporation of cryptic exons to their mRNAs, resulting in decreased levels of full-length transcripts (Fig. 1a,b) [16, 42, 57, 62]. Titration of TDP-43 suppression via a small-interfering RNA (siRNA) dose response achieved a nearly continuous range of TDP-43 expression from 100 to 18% of the level in cells treated with a control siRNA (Fig. 1c,d). This approach enabled direct comparison of relative rates of cryptic exon incorporation into UNC13A and STMN2 transcripts and revealed that STMN2 pre-mRNA is more sensitive to TDP-43 disruption than UNC13A (Fig. 1e–g). Indeed, misprocessing of STMN2 RNA and significant reduction of full-length STMN2 were detected following only 33% reduction of TDP-43 level (Fig. 1d–f; 0.125 pmol siRNA leading to 67% of normal TDP-43 levels). In contrast, UNC13A cryptic exon was only detected following 60% reduction of TDP-43 and was associated with significant loss of full-length UNC13A only after 80% of TDP-43 loss (Fig. 1d,f,g; 0.75 and 25 pmol of siRNA leading to 40% and 18% of normal TDP-43 levels, respectively). Five days after treating cells with the highest dose of siRNA (25 pmol), 70% reduction of TDP-43 led to 60% reduction in full length STMN2 but only 30% reduction in UNC13A (Sup. Fig. S1). Notably, treatment with 25 pmol of siTDP-43 was not associated with cellular toxicity (Sup. Fig S2a). Consistently, we detected misprocessing of STMN2 (Sup. Fig. S2b) but not UNC13A (Sup. Fig. S2c) in neuroblastoma SH-SY5Y cells homozygous for a N352S mutation in TARDBP which results in a decrease of TDP-43 mRNA level to around 60% of wild-type cells (Sup. Fig. S2d) [62]. Analysis of previously published RNA sequencing datasets [44, 55, 62] corroborated these results (Sup. Fig. S3). Indeed, SH-SY5Y cells with TDP-43 downregulation to 25% of control cells [62], yielded a 85% reduction of full-length STMN2; whereas, full-length UNC13A was only reduced by 20% (Sup. Fig. S3a). Notably, this relative resistance of UNC13A to TDP-43 loss occurs in SH-SY5Y cells while sequencing of this line revealed homozygosity for the ALS-associated risk allele (G) at the SNP rs12973192 that leads to an increased likelihood of UNC13A cryptic exon incorporation [16, 57]. In addition, laser-microdissected motor neurons of sporadic ALS patients [44] have significantly reduced full-length STMN2 transcripts [62] while UNC13A RNA levels were not significantly impacted (Sup. Fig. S3b). Finally, while both STMN2 and UNC13A cryptic exons were detected in FAC sorted cortical neurons lacking nuclear TDP-43 [16, 55, 57], only full-length STMN2 transcripts were significantly reduced (Sup. Fig. S3c, left panel) while UNC13A RNA levels were not significantly impacted (Sup. Fig. S3c, right panel). Collectively, these results indicate that while STMN2 and UNC13A pre-mRNAs are both misprocessed upon TDP-43 loss, STMN2 is more sensitive to mild TDP-43 disruption than UNC13A.

Several factors may underly the different sensitivities of STMN2 and UNC13A to TDP-43 loss-of-function including the binding patterns of TDP-43 on the pre-mRNAs and the level of expression of both transcripts. Analysis of a previously published individual-nucleotide resolution cross linking and immunoprecipitation (iCLIP) dataset in SH-SY5Y cells [90] identified a specific binding of TDP-43 on STMN2 cryptic exon while no read mapped in the region surrounding the UNC13A cryptic exon (Sup. Fig. S4a,b). This result may be due to the low level of UNC13A expression compared to STMN2 in SH-SY5Y cells (200 STMN2 transcripts per million compared to 20 UNC13A transcripts per million in control SH-SY5Y cells, Sup. Fig. S3a). Notably, iCLIP from SH-SY5Y cells overexpressing UNC13A minigenes demonstrated TDP-43 binding on a large region of intron 20 spanning approximately 800 bp [16] (Fig. S5a). Consistently, the region bound by TDP-43 is highly UG-rich with not only UG repeats but also motifs interrupted by adenines that were previously identified to be bound by TDP-43 [16, 74, 90]. This binding pattern strikingly contrasts with the discreet binding site of TDP-43 on STMN2 that spans only 24 nucleotides [10, 62] (Fig. S5b), likely contributing to the high sensitivity of STMN2 transcripts to TDP-43 loss.

It is also possible that other RNA binding proteins influence TDP-43-dependent cryptic splicing events. Indeed, overexpression of the heterogeneous nuclear ribonucleoprotein L (hnRNP L) was recently shown to mitigate the incorporation of cryptic exon in UNC13A upon TDP-43 loss [43]. We used genetically modified SH-SY5Y cells over-expressing either hnRNP L or a GFP control to determine whether hnRNP L also represses the incorporation of STMN2 cryptic exon upon siRNA-mediated reduction of TDP-43 (Fig. 1h). hnRNP L expression was not altered by TDP-43 loss (Fig. 1i), but as previously reported [43], significantly repressed the aberrant incorporation of UNC13A cryptic exon and increased the level of full length UNC13A (Fig. 1j). In contrast, both full length and cryptic exon-containing STMN2 transcripts were increased by overexpression of hnRNP L (Fig. 1k).

Aberrant accumulations of TDP-43 are found in approximately half of patients with Alzheimer’s disease, with the amygdala and entorhinal cortex, two anatomically adjacent regions, being the most frequently affected areas [5, 34]. To test whether TDP-43 pathology in Alzheimer’s disease patients leads to RNA mis-processing of STMN2 and UNC13A, we performed qRT-PCR using RNA from the amygdala and entorhinal cortex of 38 Alzheimer’s disease patients and 14 age-matched control individuals without significant confounding neuropathology (Sup. Table 2). Abnormal STMN2 transcripts were detected in the amygdala of 20 out of 38 patients and in the entorhinal cortex of 13 out of 27 patients with Alzheimer’s disease, and none of the control individuals (Sup. Fig. S6a–c and Sup. Table 3). We also tested misprocessing of UNC13A RNA in these areas and detected UNC13A cryptic exon in 19/38 amygdala samples and 13/27 entorhinal cortex samples from Alzheimer’s disease patients and at very low levels (qRT-PCR cycle thresholds Cts above 35 cycles) in the amygdala of 3 control individuals (Sup. Fig. S6d–f and Sup. Table 3). We then performed immunostaining to test the presence of hyperphosphorylated TDP-43 aggregates in the amygdala of these patients, and determined that the detection of STMN2 and UNC13A cryptic exons correlated with the presence of TDP-43 pathology (Fig. 2a–e and Sup. Table 3). Indeed, STMN2 cryptic exon was detected in the amygdala of 18 out of 21 Alzheimer’s disease patients positive for phosphorylated TDP-43 and was detected at low levels (Cts above 37; Sup. Fig. S6f) in only 2 out of 17 patients without detectable phosphorylated TDP-43 aggregates (Fig. 2a,b and Sup. Table 3). UNC13A cryptic exon was identified in the amygdala of 15 out of 21 Alzheimer’s disease patients with phosphorylated TDP-43 aggregates and only 4 patients without detected aggregates (Fig. 2a,d and Sup. Table 3). Similar results were obtained in the entorhinal cortex (Fig. 2c,e and Sup. Table 3). Notably, we observed a correlation between the levels of STMN2 and UNC13A cryptic exons detected in the amygdala of Alzheimer’s disease patients (Fig. 2f, Spearman r = 0.78, p < 0.0001), strengthening the link between detection of cryptic exons and loss of TDP-43 function. Finally, our observations were confirmed by RT-PCR showing the amplification of cryptic exons of both STMN2 and UNC13A transcripts in the amygdala of Alzheimer’s disease patients with TDP-43 pathology, but not in the controls or Alzheimer’s disease patients without TDP-43 pathology (Fig. 2g).

TDP-43 pathology follows a stereotypic spread from amygdala to other brain areas as Alzheimer’s disease progress [34, 67]. We classified postmortem tissues according to the presence of TDP-43 proteinopathy in the amygdala (stage I), hippocampus (stage II) and frontal cortex (stage III) following recent recommendations [67]. Staging was possible in 14 TDP-43 positive patients for which tissues were available from all three brain regions (Sup. Table 3). Only 4 out of 14 patients displayed phosphorylated TDP-43 aggregates in the frontal cortex (stage 3). In these cases, only a scarce number of inclusions were observed in the frontal cortex compared to amygdala and entorhinal cortex (Sup. Fig. S7). Consistently, STMN2 cryptic exon was only rarely detected in frontal cortex (1/19), occipital cortex (1/19) and cerebellum (4/19) in Alzheimer’s disease patients (Sup. Fig. S6a and Sup. Table 3). Notably, all the patients with cryptic exons detected in frontal cortex, occipital cortex and cerebellum also had cryptic exons in the amygdala (Sup. Table 3).

Since Alzheimer’s disease neuropathology is characterized by the accumulation of the two canonical proteins, amyloid-β and tau [14], we used semi-automated quantification to determine the area occupied by pathological inclusions of phosphorylated TDP-43, tau and amyloid-β in post-mortem amygdala of Alzheimer’s disease patients with (CE +) or without (CE−) STMN2 or UNC13A cryptic exons (Fig. 3). TDP-43 pathology is characterized by both cytoplasmic inclusions and loss of nuclear TDP-43 that appears to be an early pathogenic mechanism in TDP-43 proteinopathies [64, 68, 96]. Immunofluorescence staining confirmed that cells containing phosphorylated TDP-43 cytoplasmic inclusions did have nuclear clearance of TDP-43 consistent with TDP43 loss of function in regulating splicing, while neighbouring cells without inclusions had normal nuclear TDP-43 (Fig. 3a). Quantification of TDP-43 nuclear loss is challenging compared to quantification of phosphorylated TDP-43 aggregates [5, 7, 13, 21, 29, 34, 35, 54, 92], hence we have correlated the presence of cryptic exons with the burden of phopho-TDP-43 pathology (Fig. 3b,c). As expected, there were higher levels of phosphorylated TDP-43, tau and amyloid-β accumulations in Alzheimer’s disease patients compared to control individuals (Fig. 3b–e). However, while the pathological burden in tau (Fig. 3d) or amyloid-β (Fig. 3e) were not significantly different in tissues with or without cryptic exons, we observed higher pathological burden of TDP-43 in patients with accumulation of STMN2 or UNC13A cryptic exons (Fig. 3c). Notably, the burden of TDP-43 pathology (percentage of area positive for TDP-43) positively correlated with the levels of both STMN2 (Spearman r = 0.52; p-value = 0.006) and UNC13A (Spearman r = 0.56; p-value = 0.002) cryptic exons in the amygdala. Collectively these results demonstrate that mis-processing of STMN2 and UNC13A RNAs is a pathological hallmark of TDP-43-associated Alzheimer’s disease, with amygdala and entorhinal cortex being the most affected areas.

We analyzed STMN2 and UNC13A transcripts in bulk RNA-seq from post-mortem brain regions of control and Alzheimer’s disease patients from data available through the Mount Sinai/JJ Peters VA Medical Center Brain Bank (MSBB-AD) [97] and an independent dataset generated at Mayo Clinic [2, 3, 98]. In the Mount Sinai dataset, RNA-seq of the parahippocampal gyrus, inferior frontal gyrus, frontal pole, and superior temporal gyrus from more than 40 control individuals and 95 Alzheimer’s disease patients were analyzed (Fig. 4a–c, Sup. Figs. S8, S9). Of note, no information was available concerning the presence or absence of TDP-43 pathology in this cohort. However, we observed a significant increase of truncated STMN2 in the inferior frontal gyrus (Fig. 4a) of Alzheimer’s disease patients; whereas, full-length STMN2 RNA levels were significantly decreased in Alzheimer’s disease patients compared to controls, in all the regions tested (Fig. 4b and Sup. Fig. S8b). While the number of reads mapping to UNC13A cryptic exon was extremely low in all samples, a significant decrease in UNC13A full-length was detected in the inferior frontal gyrus, parahippocampal gyrus and superior temporal gyrus from Alzheimer’s disease patients compared to control individuals (Fig. 4c and Sup. Fig. S8c). The Mayo Clinic dataset [2, 3, 98] consisted of bulk RNA-seq of the temporal cortex (superior temporal gyrus) from 69 control individuals and 80 Alzheimer’s disease patients. Here, TDP-43 pathology was assayed in 38 of the Alzheimer’s disease patients with 18 showing accumulation of phosphorylated TDP-43, consistent with previous reports of approximately 50% of Alzheimer’s disease patients having TDP-43 pathology [5, 7, 13, 21, 29, 34, 35, 54, 92]. Analysis of the Mayo Clinic dataset showed a significant enrichment of STMN2 cryptic exon in the temporal cortex of Alzheimer’s disease patients compared to controls (Fig. 4d) and a significant decrease in full-length STMN2 only in Alzheimer’s disease patients with TDP-43 pathology compared to controls (Fig. 4e). UNC13A full-length was significantly decreased in patients with TDP-43 pathology (Fig. 4f). We further explored this dataset by subdividing Alzheimer’s disease cases (including cases with and without TDP-43 pathology) according to their Braak stages, a method to classify the evolution of the disease by evaluating the spread of tau pathology in Alzheimer’s disease brain [14, 15]. This analysis revealed a significant increase of STMN2 cryptic exon in the more advanced Braak stages (Braak stages V and VI), a trend for decrease in full-length STMN2 in the Mayo Clinic dataset, and a significant decrease in the full-length transcripts of STMN2 and UNC13A in the Mount Sinai dataset (Fig. 4g and Sup. Fig. S9). In the cases from Mayo Clinic with Braak stages V and VI for which the TDP-43 status was available, we observed a significant increase of STMN2 cryptic exon in the temporal cortex of patients with Braak stage VI and TDP-43 pathology (Sup. Fig. 10a). Full-length STMN2 and UNC13A transcripts were also significantly reduced in both Braak stages V and VI when comparing brains with and without phosphorylated TDP-43 (Sup. Fig. 10).

Collectively this analysis establishes, in a large sample population, that STMN2 and UNC13A transcripts are altered in Alzheimer’s disease patients with advanced Braak stages when the burden of TDP-43 pathology is high.

The authors report no competing interests to declare.

The data from this study are available from the corresponding authors, upon request. Mayo Clinic brain RNA-seq data are available from the AD Knowledge Portal (syn5550404). MSBB RNA-seq data are available from the AD Knowledge Portal (syn3157743).